custom-tool-development

You create custom tools by defining functions with the @tool_define decorator in a Python file within the .memstack/tools/ directory, specifying parameters using JSON Schema for the agent to discover and use.

You test custom agent tools in isolation by using the CustomToolLoader methodology, which allows you to verify individual tool execution and parameter handling independently before integrating them fully into the agent workflow.

Custom tool import errors often occur due to missing @tool_define decorators, incorrect file structure conventions outside the .memstack/tools/ directory, or naming conflicts caused by duplicate tool names across files.

You handle errors and return results from a custom agent tool by utilizing the ToolResult object within the function logic, while accessing necessary runtime information through the ToolContext object.

You can integrate a proprietary internal API by wrapping the API calls inside a custom tool definition, configuring the required parameters with JSON Schema, and ensuring the tool is testable by the agent.

The ToolContext provides essential runtime information and environment access to your custom tool during execution, allowing the function to interact safely with the agent platform and return structured ToolResult outputs.